How can a bank or credit union get recommended by ChatGPT?

The short version

A bank or credit union gets recommended by ChatGPT when the model can read its site, quote its facts, and match them to the question a person asked. You cannot force a recommendation. You can remove the reasons your institution is invisible, and you can make the facts that would earn a mention easy to retrieve. Here is what to check first.

Start with the crawler

ChatGPT search uses OAI-SearchBot to surface websites in its answers, according to OpenAI's overview of its crawlers. If your site blocks that bot, it will not be shown in ChatGPT search answers, though it can still appear as a navigational link. OpenAI recommends allowing OAI-SearchBot in robots.txt and permitting requests from its published IP ranges.

Two practical points. First, OAI-SearchBot and GPTBot serve different purposes: one is search, one is training. You can permit search crawling while disallowing training crawling, so a cautious stance on training does not have to mean total absence from search. Second, check with whoever manages your robots.txt and any firewall or bot-filtering rules. Financial sites often block aggressive crawler filtering by default, and a block at the CDN level does the same damage as a block in robots.txt.

Make the facts quotable

Access is necessary but not sufficient. When a person asks ChatGPT which credit union has a low-APR auto loan in a specific county, the engine builds its answer from text it can quote. That means:

Rates, fees, and eligibility written as plain HTML text, not images, not tables flattened into screenshots, not buried in PDFs.

Effective dates on every rate and offer. An answer engine hesitates to cite a rate with no date on it.

Location and eligibility language stated directly: which counties you serve, who can join, what the minimums are.

Product pages that stand alone. A page that answers "what is your current CD rate and term" without requiring a click through a rate disclosure tool is far easier to quote.

Rates, fees, and disclosures are regulated content, so give those pages a careful eye when you rewrite them for clarity. Clear and compliant are not in conflict; a dated, plainly worded rate table serves both.

Test with real prompts

The only measurement that matters is what ChatGPT actually says. Ask the questions your prospective members ask, in the phrasing they use. Not "best banking products" but "which bank in Scanley County has no monthly checking fee." Run the same prompt across sessions. Note whether you are named, who is named instead, and what source the answer cites. That last detail is the most useful: the citation tells you which page the engine found quotable, which tells you what to build.

Your rankings in traditional search can look fine while ChatGPT skips you entirely, because the answer is assembled from a different retrieval path. There is no alert when it happens. Checking prompts by hand works at first; when the question list grows past what one marketer can track, Scanley measures this across the AI engines on a schedule.

What to do this month

Verify OAI-SearchBot is allowed in robots.txt and not blocked at the firewall or CDN.

Decide separately on GPTBot, since search access and training access are distinct choices.

Pick your three most competitive products and make sure each has a dated, plain-text page a crawler can quote.

Run ten real prompts and record who gets named and which sources get cited.

Repeat monthly. Change one thing at a time so you can attribute movement.

If you are a community institution, the specifics differ: community banks and credit unions tend to win on local specificity, which is exactly the kind of fact an answer engine can quote if you write it down.

Questions people ask